Using the Bayesian updating approach to improve the spatial and temporal transferability of real-time crash risk prediction models

This study aimed to improve the spatial and temporal transferability of the real-time crash risk prediction models by using the Bayesian updating approach. Data from California’s I-880N freeway in 2002 and 2009 and the I-5N freeway in 2009 were used. The crash risk models for these three datasets are quite different from each other. The model parameters do not remain stable over time or space. The transferability evaluation results show that the crash risk models cannot be directly transferred across time and space. The updating results indicate that the Bayesian updating approach is effective in improving both spatial and temporal transferability even when new data are limited. The predictive performance of the updated model increases with an increase in the sample size of the new data. In addition, when limited new data are available, updating an existing model is better than developing a model using the limited new data.     

Further evidence on the temporal stability of disaggregate travel demand models

Current evidence on the transferability of disaggregate travel demand models is inconclusive. Adding to this body of research, the present analysis focuses upon the temporal characteristics of work trip behavior in the San Francisco Bay Area. Using before and after data sets associated with the BART Impact Travel Study, multinomial logit models of work trip modal choice are estimated. The results indicate that the general form and the coefficient estimates of a pre BART model are transferable in time. Moreover, when updated to reflect BART's presence, the model's predictive success and its implied elasticity measures are generally accurate, relative to those implied by reestimating the entire model on post BART data. Finally, as economic theory would predict, elasticity measures of the service related variables were found to increase over time.     

Travel demand forecasts improved by using cross-sectional data from multiple time points

Forecasts of travel demand are often based on data from the most recent time point, even when cross-sectional data is available from multiple time points. This is because forecasting models with similar contexts have higher transferability, and the context of the most recent time point is believed to be the most similar to the context of a future time point. In this paper, the author proposes a method for improving the forecasting performance of disaggregate travel demand models by utilising not only the most recent dataset but also an older dataset. The author assumes that the parameters are functions of time, which means that future parameter values can be forecast. These forecast parameters are then used for travel demand forecasting. This paper describes a case study of journeys to work mode choice analysis in Nagoya, Japan, using data collected in 1971, 1981, 1991, and 2001. Behaviours in 2001 are forecast using a model with only the most recent 1991 dataset and models that combine the 1971, 1981, and 1991 datasets. The models proposed by the author using data from three time points can provide better forecasts. This paper also discusses the functional forms for expressing parameter changes and questions the temporal transferability of not only alternative-specific constants but also level-of-service and socio-economic parameters.     

Comparison of updating techniques in transferability analysis of work trip mode choice models in developing countries

This study analyzes the performances of updating techniques in transferability of mode choice models in developing countries. A model specification, estimated in Ho Chi Minh City, was transferred to Phnom Penh. Naïve transfer and four updating methods associated with small sized samples were used in the transfer process and were evaluated based on statistical perspective and predictive ability. The study also illustrates the problems faced in model transferability development, due to the lack of available and suitable data in Phnom Penh. This lack is strongly related to different methods and structures applied in collecting the data. Simplified approaches to the difficulties are proposed in the study. The results show that updating ASCs, updating both ASCs and scale parameter, and use of combined transfer estimators all produce significant improvement, both statistically and in predictability, in updating the model. The last two methods have proven to be superior to the first method, owing to the inclusion of transfer bias considerations in the estimations. However, small data samples should not have large transfer bias when using combined transfer estimators. It is also concluded that naïvely transferring a model is not recommended, and Bayesian updating should be avoided when transfer bias exists. Copyright © 2010 John Wiley & Sons, Ltd.     

Spatial Transferability and Updating Analysis of Mode Choice Models in Developing Countries

ABSTRACT

This article examines the spatial transferability of mode choice models in developing countries. An evaluation of the updating procedure and sample size are also included in the study. Because of the insufficiency of model coefficients in explaining differences in unmeasured modal attributes, naïvely transferring a model is not recommended. An understanding of the transport characteristics in both the estimation context and the application context is required, in order to justify whether a variable is transferable or not. Four updating procedures – updating alternative specific constants (ASCs), updating ASCs and scale parameter, the combined transfer estimator and Bayesian updating associated with three sets of small sample sizes – are applied to improve transferability. In general, the first three approaches produce significant improvements. It is also proposed that a minimum small sample size of 400 observations is necessary for updating purposes.     

Home-based trip end models — A comparison between category analysis and regression analysis procedures

The trip end models which have been used in past transportation studies are briefly summarised. Problems associated with the use of zone-based models are outlined and reasons are given to support the development of models at the household rather than zonal level.It is suggested that recent developments which have taken place in household-based models have not been entirely logical. In particular, arguments between regression models and category analysis models have been confused with the use of aggregate (zonal) as against disaggregate (household) data — regression models being associated with the use of zonal data and category analysis models with household data. Misunderstood arguments and false notions regarding sample sizes have directed attention from the regression analysis approach.A detailed comparison of the category analysis and regression analysis methods for developing household-based trip end models is given. Both methods have been applied using data from the Monmouthshire Land Use Transportation Study. The regression results reported are from a very preliminary analysis and contain a number of anomalies, although it is thought that sufficient work has been done to provide an objective evaluation of the two methods.It is recommended that the household regression approach should be further investigated since it has advantages as a modelbuilding procedure and makes better use of sample data. A certain amount of categorisation of household types is necessary and the investigations would attempt to determine the best balance between categorisation and regression fitting. Further development will be restricted if the trend towards minimum sample sizes of about 1000 households is continued. Larger samples should be taken in certain circumstances to pursue development work.     

Flexible long range planning using low cost information

Contemporary transport planning requires a flexible modelling approach which can be used to monitor the implementation of a long term plan checking regularly its short term performance with easily available data; the original model is periodically updated using low cost information and this allows the evaluation of the changes to the plan which may be required. Such an approach requires models suited to regular updating and to the use of data from different sources. Models to update trip matrices from traffic counts have been available for some time; however, the estimation and/or updating of other model stages with low cost data has escaped analytical treatment. The paper discusses this idea and formulates the updating problem for an example involving a joint destination/mode choice model under various assumptions about the nature of the available data. Analytical solutions are proposed as well as some general conclusions.requests for offprints     

Synthetic household travel survey data simulation

Due to the high cost, low response rate and time-consuming data processing, few Metropolitan Planning Organizations can afford collecting household travel survey data as frequently as needed. This paper presents a methodology to simulate disaggregate and synthetic household travel survey data by examining the feasibility of the spatial transferability of travel data. Households are clustered into several homogeneous groups to identify the distributions of their travel attributes. These distributions are then transferred to similar groups in other regions. Furthermore, updating methods are suggested and developed to calibrate the parameters of the transferred distributions for the application area. A user friendly software is developed that facilitates the entire process. To validate the model, a synthetic population for the state of New York, excluding the New York City, is generated by a two-stage population synthesis procedure. Then, travel attributes of each household are simulated and by linking the generated travel data to the synthetic population, a synthetic household travel dataset is generated for the application context. Finally, using a new validation dataset from the application area, comparisons against the simulated data are made to examine the effectiveness of the simulation process.     

A comparative study of aggregate and disaggregate gravity models using Seoul metropolitan subway trip data

Abstract

Existing origin constrained and doubly constrained gravity models have not been compared, theoretically or empirically, in terms of their forecasting power. Due to the newly advanced technology of intelligent transport systems, the expanded data presently available have made various models more comparable in terms of forecasting power. This paper uses archived automatic passenger counting (APC) data for urban rail in the Seoul metropolitan area. The APC data contains information about each trip's origin, destination, ticket type, fare, and distance on a daily basis. The objective of this paper is to compare the goodness-of-fit of aggregate and disaggregate gravity modeling using these data. A Hyman aggregate gravity model is used as the aggregate model without the spatial effect. The disaggregate model adopts a multinomial logit as the destination choice model with the spatial effect. In general, while the formulation of aggregate and disaggregate gravity model models are similar, the calibration and parameter estimation methods of the two models are different. As a result, this empirical study demonstrates that the variation in goodness-of-fit and forecasting power largely depends on the estimation method and selected variables. The forecasting power of the disaggregate modeling approach outperforms that of the aggregate model. This paper further confirms that spatial arrangement plays important roles in gravity modeling.     

Individual trip rate transferability analysis based on a decision tree approach

Transferring trip rates to areas without local survey data is a common practice which is typically performed in an ad hoc fashion using household-based cross-classification tables. This paper applies a rule-based decision tree method to develop individual-level trip generation models for eight different trip purposes as defined in the US National Household Travel Survey in addition to daily vehicle miles traveled. For each trip purpose, the models are obtained by finding the best fitted statistical distribution to each of the final decision tree clusters while considering the correlation between the trip rates for other trip purposes. The rule-based models are sensitive to changes in demographics. The performance of the models is then tested and validated in a transferability application to the Phoenix Metropolitan Region. These models can be employed in a disaggregate microsimulation framework to generate trips with different purposes at the individual or household level.     

Adaptive traffic parameter prediction: Effect of number of states and transferability of models

Traffic parameters can show shifts due to factors such as weather, accidents, and driving characteristics. This study develops a model for predicting traffic speeds under these abrupt changes within regime switching framework. The proposed approach utilizes Hidden Markov, Expectation Maximization, Recursive Least Squares Filtering, and ARIMA methods for an adaptive forecasting method. The method is compared with naive and mean updating linear and nonlinear time series models. The model is fitted and tested extensively using 1993 I-880 loop data from California and January 2014 INRIX data from Virginia. Analysis for number of states, impact of number of states on forecasting, prediction scope, and transferability of the model to different locations are investigated. A 5-state model is found to be providing best results. Developed model is tested for 1-step to 45-step forecasts. The accuracy of predictions are improved until 15-step over nonadaptive and mean adaptive models. Except 1-step predictions, the model is found to be transferable to different locations. Even if the developed model is not retrained on different datasets, it is able to provide better or close results with nonadaptive and adaptive models that are retrained on the corresponding dataset.     

Congestion and safety: A spatial analysis of London

A disaggregate spatial analysis, using enumeration district data for London was conducted with the aim of examining how congestion may affect traffic safety. It has been hypothesized that while congested traffic conditions may increase the number of vehicle crashes and interactions, their severity is normally lower than crashes under uncongested free flowing conditions. This is primarily due to the slower speeds of vehicles when congestion is present. Our analysis uses negative binomial count models to examine whether factors affecting casualties (fatalities, serious injuries and slight injuries) differed during congested time periods as opposed to uncongested time periods. We also controlled for congestion spatially using a number of proxy variables and estimated pedestrian casualty models since a large proportion of London casualties are pedestrians. Results are not conclusive. Our results suggest that road infrastructure effects may interact with congestion levels such that in London any spatial differences are largely mitigated. Some small differences are seen between the models for congested versus uncongested time periods, but no conclusive trends can be found. Our results lead us to suspect that congestion as a mitigator of crash severity is less likely to occur in urban conditions, but may still be a factor on higher speed roads and motorways.     

Freight transportation demand: A survey of recent econometric studies

This paper surveys econometric studies of freight transportation demand which have been published since the mid-1970s. It describes the variables, data sources, and estimation procedures utilized by the studies. In addition, it summarizes their statistical results. The studies included in this survey typically accounted for freight rates and service characteristics (e.g., transit time and reliability). Data sources often varied across the studies.Based on the data they utilized, the surveyed studies are classified as either aggregate or disaggregate. The data in the aggregate studies consist of information on total flows by modes at the regional or national level, while the data in the disaggregate studies pertain to individual shipments. The earlier aggregate studies estimated linear logit models. It has been pointed out that when they are estimated on aggregate data these models are subject to certain shortcomings. To avoid these shortcomings, more recent aggregate studies have estimated flexible forms such as translog functions. The disaggregate studies surveyed in this paper used either logit or probit models.Statistical results often varied with the commodities analyzed, making it somewhat difficult to generalize the findings of the different studies. One finding common to several studies is that freight rates have a significant impact on shipment decisions. This paper discusses certain theoretical and empirical limitations of the surveyed studies. It also offers suggestions for future research in freight transport demand. Freight demand models can be used to examine various effects of the recent deregulation in freight transportation.     

Bayesian updating of trip generation data: Combining national trip generation rates with local data

With traffic impact analyses and impact fee assessment becoming more popular, the need for accurately estimating the trip generation rate of a proposed development is becoming more important. An overwhelming percentage of state transportation agencies depend either partly or entirely on the ITETrip Generation Report to predict the traffic that will be attracted to and/or produced from a proposed development. However, the rates obtained from the ITE publication have been derived from data collected throughout the United States. They represent a national average and fail to take into account the local trip generation characteristics that the site under consideration might have. This paper establishes a methodology for obtaining more reliable local trip generation rates using Bayesian statistics. In this method, the ITE rates are assumed to be the prior information, which are updated using limited local trip generation data that are available. The method also allows for temporal updating, incorporating subjective judgment and using borrowed data in the updating procedure. Sample calculations in this paper illustrate the developed methodology.     

Modeling interdependencies between vehicle transaction,residential relocation and job change

This paper introduces a vehicle transaction timing model which is conditional on household residential and job relocation timings. Further, the household residential location and members’ job relocation timing decisions are jointly estimated. Some researchers have modeled the household vehicle ownership decision jointly with other household decisions like vehicle type choice or VMT; however, these models were basically static and changes in household taste over time has been ignored in nearly all of these models. The proposed model is a dynamic joint model in which the effects of land-use, economy and disaggregate travel activity attributes on the major household decisions; residential location and members’ job relocation timing decisions for wife and husband of the household, are estimated. Each of these models is estimated using both the Weibull and log-logistic baseline hazard functions to assess the usefulness of a non-monotonic rather than monotonic baseline hazard function. The last three waves of the Puget Sound Panel Survey data and land-use, transportation, and built environment variables from the Seattle Metropolitan Area are used in this study as these waves include useful explanatory variables like household tenure that were not included in the previous waves.     

Evaluation of transfer methods for spatial travel demand models

The main goal of the research was to compare alternative methods of spatial transfer as a function of sample size. The study was based on the mobility surveys conducted in the Helsinki Metropolitan Area in 1995 and in the Turku region in 1997. The Helsinki Metropolitan Area data base was used to estimate the models that were to be transferred. The data base in the Turku region represents the application context to which the estimated Helsinki Metropolitan Area models are transferred. The transfer procedures examined were Bayesian updating, combined transfer estimation, transfer scaling, and joint context estimation procedures. To explore the impact of sample size on transferring performance, model transferability was tested using six different sample sizes. The model transferability was examined by comparing the transferred models to the models estimated using the entire set of the data which can be regarded as the best estimate representing “the real situation”. The results indicated that joint context estimation gives the best prediction performance in almost all cases. In particular, the method is useful if the difference in the true parameters between the two contexts is large or only some of the model coefficients are precise. The applicability of joint context estimation can be improved by viewing the coefficients as variable-oriented and emphasizing precise and imprecise coefficients differently.     

Fundamentals of discrete multimodal choice modelling

Abstract

In this paper an overview is given of the most relevant issues relating to the application of multimodal choice models, with particular emphasis on disaggregate modal split models. The paper considers questions of data, such as type of data, alternative sampling strategies and problems of measurement; and modelling issues, such as model specification and estimation, including a good presentation of the statistical techniques'available. The paper also addresses the aggregation problem, which lies at the heart of one of today's most hotly contested debates: whether to use aggregate or disaggregate models for policy analysis, and in which circumstances.     

Updating predictive accident models of modern rural single carriageway A-roads

Abstract

Reliable predictive accident models (PAMs) are essential to design and maintain safe road networks, and yet the models most commonly used in the UK were derived using data collected 20 to 30 years ago. Given that the national personal injury accident total fell by some 30% in the last 25 years, while road traffic increased by over 60%, significant errors in scheme appraisal and evaluation based on the models currently in use seem inevitable. In this paper, the temporal transferability of PAMs for modern rural single carriageway A-roads is investigated, and their predictive performance is evaluated against a recent data set. Despite the age of these models, the PAMs for predicting the total accidents provide a remarkably good fit to recent data and these are more accurate than models where accidents are disaggregated by type. The performance of the models can be improved by calibrating them against recent data.     

Scrutinizing individuals’ leisure-shopping travel decisions to appraise activity-based models of travel demand

Activity-based models for modeling individuals’ travel demand have come to a new era in addressing individuals’ and households’ travel behavior on a disaggregate level. Quantitative data are mainly used in this domain to enable a realistic representation of individual choices and a true assessment of the impact of different Travel Demand Management measures. However, qualitative approaches in data collection are believed to be able to capture aspects of individuals’ travel behavior that cannot be obtained using quantitative studies, such as detailed decision making process information. Therefore, qualitative methods may deepen the insight into human’s travel behavior from an agent-based perspective. This paper reports on the application of a qualitative semi-structured interview method, namely the Causal Network Elicitation Technique (CNET), for eliciting individuals’ thoughts regarding fun-shopping related travel decisions, i.e. timing, shopping location and transport mode choices. The CNET protocol encourages participants to think aloud about their considerations when making decisions. These different elicited aspects are linked with causal relationships and thus, individuals’ mental representations of the task at hand are recorded. This protocol is tested in the city centre of Hasselt in Belgium, using 26 young adults as respondents. Response data are used to apply the Association Rules, a fairly common technique in machine learning. Results highlight different interrelated contexts, instruments and values considered when planning a trip. These findings can give feedback to current AB models to raise their behavioral realism and to improve modeling accuracy.     

Children’s mode choice for the school trip: the role of distance and school location in walking to school

Rising levels of childhood obesity in the United States and a 75% decline in the proportion of children walking to school in the past 30 years have focused attention on school travel. This paper uses data from the US Department of Transportation’s 2001 National Household Travel Survey to analyze the factors affecting mode choice for elementary and middle school children. The analysis shows that walk travel time is the most policy-relevant factor affecting the decision to walk to school with an estimated direct elasticity of −0.75. If policymakers want to increase walking rates, these findings suggest that current policies, such as Safe Routes to School, which do not affect the spatial distribution of schools and residences will not be enough to change travel behavior. The final part of the paper uses the mode choice model to test how a land use strategy—community schools—might affect walking to school. The results show that community schools have the potential to increase walking rates but would require large changes from current land use, school, and transportation planning practices.